ENGINEERING THE PUBLIC: BIG DATA, SURVEILLANCE AND COMPUTATIONAL POLITICS
Forthcoming: July 2014 issue of First Monday, Volume 19, Number 7
1
Engineering the Public: Big Data, Surveillance and Computational Politics
By Zeynep Tufekci
Abstract: Digital technologies have given rise to a new combination of big data and
computational practices which allow for massive, latent data collection and sophisticated
computational modeling, increasing the capacity of those with resources and access to use these
tools to carry out highly effective, opaque and unaccountable campaigns of persuasion and social
engineering in political, civic and commercial spheres. I examine six intertwined dynamics that
pertain to the rise of computational politics: the rise of big data, the shift away from
demographics to individualized targeting, the opacity and power of computational modeling, the
use of persuasive behavioral science, digital media enabling dynamic real-time experimentation,
and the growth of new power brokers who own the data or social media environments. I then
examine the consequences of these new mechanisms on the public sphere and political
campaigns.
ENGINEERING THE PUBLIC: BIG DATA, SURVEILLANCE AND COMPUTATIONAL POLITICS
Forthcoming: July 2014 issue of First Monday, Volume 19, Number 7
2
Contents
Introduction
Engineering the Public: From Broadcast to the Internet
New Dynamics of Persuasion, Surveillance, Campaigning and Social Engineering
1. Big Data
2. Emergent Computational Methods
3. Modeling
4. Behavior Science
5. Experimental Science in Real Time Environments:
6. Platforms and Algorithmic Governance
Consequences and Power of Big Data Analytics
Discussion and Conclusion
ENGINEERING THE PUBLIC: BIG DATA, SURVEILLANCE AND COMPUTATIONAL POLITICS
Forthcoming: July 2014 issue of First Monday, Volume 19, Number 7
3
Introduction: Engineering the Public: Big Data, Surveillance and Computational Politics
Emergence of networked technologies instilled hopes that interactivity in the public
sphere could help limit, or even cure, some of the ailments of late modern democracies. In
contrast to broadcast technologies, the Internet offers expansive possibilities for horizontal
communication among citizens, while drastically lowering the costs of organizing and access to
information (Shirky, 2008). Indeed, the Internet has been a critical tool for many social
movements (Tufekci and Freelon, 2013).
However, Internet’s propensity for citizen empowerment is neither unidirectional, nor
straightforward. The same digital technologies have also given rise to a data-analytic
environment that favors the powerful, data-rich incumbents, and the technologically adept,
especially in the context of political campaigns. These counter-trends arises specifically from an
increased exploitation on big data, that is, very large datasets of information gleaned from online
footprints and other sources, along with analytic and computational tools.
Big data is often hailed for its ability to add to our knowledge in novel ways and to enrich
our understanding (Lazer et al., 2009; Lohr, 2012). However, big data also needs to be examined
as a political process involving questions of power, transparency and surveillance. In this paper, I
argue that big data and associated new analytic tools foster more effective –and less
transparent— “engineering of consent” (Bernays, 1947) in the public sphere. As a normative
(but contested) ideal, the public sphere is envisioned by Habermas (1991) as the location and
place in which rational arguments about matters concerning the public, especially regarding
issues of governance and the civics can take place, freed from constraints of status and identity.
The public sphere should be considered at once a “normative ideal” as well as an institutional
analysis of historical practice (Calhoun, 1993). As actual practice, the public sphere pertains to
ENGINEERING THE PUBLIC: BIG DATA, SURVEILLANCE AND COMPUTATIONAL POLITICS
Forthcoming: July 2014 issue of First Monday, Volume 19, Number 7
4
“places”— intersections and commons—where these civic interactions take place, and which are
increasingly online. This shift to a partially online public sphere, which has brought about the
ability to observe, surveil and collect these interactions in large datasets, has given rise to
computational politics, the focus of this paper.
Computational politics refers applying computational methods to large datasets derived
from online and offline data sources for conducting outreach, persuasion and mobilization in the
service of electing, furthering or opposing a candidate, a policy or legislation Computational
politics is informed by behavioral sciences and refined using experimental approaches, including
online experiments, and is often used to profile voters, sometimes in the aggregate but especially
at the individual level, and to develop methods of persuasion and mobilization which, too, can be
individualized. Thus, computational politics is a set of political practices the rise of which
depends on, but is not solely defined by, the existence of big data and accompanying analytic
tools and is defined by the significant information asymmetry—the campaigns know a lot about
individual voters while voters don’t know what campaigns know about them (Federal Trade
Commission, 2014),
While computational politics in its current form includes novel applications, the historical
trends discussed in this paper predate the spread of the Internet. In fact, there was already a
significant effort underway to use big data for purposes of marketing, and the progression of
using marketing techniques for politics—and “selling of the president”—clearly reflects longer-
term trends (McGinnis, 1988). However, computational politics introduces significant qualitative
differences to that long march of historical trends. Unlike previous data collection efforts (for
example, collating magazine subscriptions or car type purchases) which required complicated,
roundabout inferences about their meaning (does a magazine subscription truly signal a voter
ENGINEERING THE PUBLIC: BIG DATA, SURVEILLANCE AND COMPUTATIONAL POLITICS
Forthcoming: July 2014 issue of First Monday, Volume 19, Number 7
5
preference?) and allowed only broad profiling in the aggregate, this data provides significantly
more individualized profiling and modeling, much greater data depth, and can be collected in an
invisible, latent manner and delivered individually.
Computational politics turns political communication into an increasingly personalized,
private transaction and thus fundamentally reshapes the public sphere, first and foremost by
making it less and less public as these approaches can be used to both profile and interact
individually with voters outside the public sphere (such a Facebook ad aimed at that particular
voter, seen only by her). Overall, the impact is not so much like increasing the power of a
magnifying glass as it is like re-purposing the glass by putting two or more together to make
fundamentally new tools, like the microscope or the telescope, turning unseen objects into
objects of scientific inquiry and manipulation.
Big data’s impact on the public sphere through computational politics operates through
multiple intertwined dynamics, and the purpose of this paper is to define, explain and explore
them both individually, and also within the context of their intertwining. First, the rise of digital
mediation of social, political and financial interactions has resulted in an exponential increase in
the amount and type of data available, especially to large organizations that can afford the
access, i.e. big data. Second, emergent computational methods allow political targeting to move
beyond aggregated group-based analysis and profiling to modeling of specific individuals.
Third, such modeling allows for acquiring answers about an individual without directly asking
questions to the individual, thus opening the door to a new wave of techniques reliant on
subterfuge and opacity. Fourth, advances in behavioral sciences have resulted in a move away
from models of the “rational human” towards more refined and realistic models of human
behavior. In combination with the other dynamics outlined here, these models allow for